Abstract: System and method for document image detection, comprising: producing, using a neural network, a superpixel segmentation map of an input image; generating a superpixel binary mask by associating each superpixel of the superpixel segmentation map with a class of a predetermined set of classes; identifying one or more connected components in the superpixel binary mask; for each connected component of the superpixel binary mask, identifying a corresponding minimum bounding polygon; creating one or more image dividing lines based on the minimum bounding polygons; and defining boundaries of one or more objects of interest based on at least a subset of the image dividing lines.

Abstract: Aspects and implementations provide for mechanisms of detection and decoding of barcodes in images. The disclosed techniques include estimating dimensions of a module of a barcode based on geometric characteristics of a barcode image, forming hypotheses that group modules into barcode symbols, and assessing viability of formed hypotheses. Various operations of the techniques may involve the use of neural networks, including estimation of module dimensions and assessment of groupings of modules into lines and lines into barcode symbols. The techniques may be used for decoding of barcodes captured in images of unfavorable conditions, including blur, perspective, sub-optimal lighting, barcode deformation, and the like. The techniques may be applied to decoding linear one-dimensional barcodes, two-dimensional barcodes, and stacked linear barcodes.

Abstract: A computer-implemented method for image capture by a mobile device, comprising: receiving, by a video capturing application running on a mobile device, a video stream from a camera of the mobile device; identifying a specific frame of the video stream; generating a plurality of hypotheses defining image borders within the specific frame; selecting, by a neural network, a particular hypothesis among the plurality of hypotheses; producing a candidate image by applying the particular hypothesis to the specific frame; determining a value of a quality metric of the candidate image; determining that the value of the quality metric of the candidate image exceeds one or more values of the quality metric of one or more previously processed images extracted from the video stream; wherein the image capture application is a zero-footprint application.

Abstract: Aspects of the disclosure provide for systems and processes for assessing image quality for optical character recognition (OCR), including but not limited to: segmenting an image into patches, providing the segmented image as an input into a first machine learning model (MLM), obtaining, using the first MLM, for each patch, first feature vectors representative of a reduction of imaging quality in a respective patch, and second feature vectors representative of a text content of the respective patch, providing to a second MLM the first feature vectors and the second feature vectors, and obtaining, using the second MLM, an indication of suitability of the image for OCR.

Abstract: A method of generating and optimizing a codebooks for document analysis comprises: receiving a first set of document images; extracting a plurality of keypoint regions from each document image of the first set of document images; calculating local descriptors for each keypoint region of the extracted keypoint regions; clustering the local descriptors such that each center of a cluster of local descriptors corresponds to a respective visual word; generating a codebook containing a set of visual words; and optimizing the codebook by maximizing mutual information (MI) between a target field of a second set of document images and at least one visual word of the set of visual words.

Abstract: Systems and methods are disclosed to receive a training data set comprising a plurality of document images, wherein each document image of the plurality of document images is associated with respective metadata identifying a document field containing a variable text; generate, by processing the plurality of document images, a first heat map represented by a data structure comprising a plurality of heat map elements corresponding to a plurality of document image pixels, wherein each heat map element stores a counter of a number of document images in which the document field contains a document image pixel associated with the heat map element; receive an input document image; and identify, within the input document image, a candidate region comprising the document field, wherein the candidate region comprises a plurality of input document image pixels corresponding to heat map elements satisfying a threshold condition.

Abstract: Aspects of the disclosure provide for mechanisms for identification of fields in documents using neural networks. A method of the disclosure includes obtaining a layout of a document, the document having a plurality of fields, identifying the document, based on the layout, as belonging to a first type of documents of a plurality of identified types of documents, identifying a plurality of symbol sequences of the document, and processing, by a processing device, the plurality of symbol sequences of the document using a first neural network associated with the first type of documents to determine an association of a first field of the plurality of fields with a first symbol sequence of the plurality of symbol sequences of the document.

Abstract: An example method of document classification comprises: detecting a set of keypoints in an input image; generating a set of keypoint vectors, wherein each keypoint vector of the set of keypoint vectors is associated with a corresponding keypoint of the set of keypoints; extracting a feature map from the input image; producing a combination of the set of keypoint vectors with the feature map; transforming the combination into a set of keypoint mapping vectors according to a predefined mapping scheme; estimating, based on the set of keypoint mapping vectors, a plurality of importance factors associated with the set of keypoints; and classifying the input image based on the set of keypoints and the plurality of importance factors.

Abstract: In one embodiment, a system receives an image depicting a line of text. The system segments the image into two or more fragment images. For each of the two or more fragment images, the system determines a first hypothesis to segment the fragment image into a first plurality of grapheme images and a first fragmentation confidence score. The system determines a second hypothesis to segment the fragment image into a second plurality of grapheme images and a second fragmentation confidence score. The system determines that the first fragmentation confidence score is greater than the second fragmentation confidence score. The system translates the first plurality of grapheme images defined by the first hypothesis to symbols. The system assembles the symbols of each fragment image to derive the line of text.

Abstract: Aspects of the disclosure provide for mechanisms for identification of table partitions in documents using neural networks. A method of the disclosure includes obtaining a plurality of symbol sequences of a document having at least one table, determining a plurality of vectors representative of symbol sequences having at least one alphanumeric character or a table graphics element, processing the plurality of vectors using a first neural network to obtain a plurality of recalculated vectors, determining an association between a first recalculated vector and a second recalculated vector, wherein the first recalculated vector is representative of an alphanumeric sequence and the second recalculated vector is associated with a table partition, and determining, based on the association between the first recalculated vector and the second recalculated vector, an association between the alphanumeric sequence and the table partition.

Abstract: Aspects of the disclosure provide for mechanisms for identification of blocks of associated words in documents using neural networks. A method of the disclosure includes obtaining a plurality of words of a document, the document having a first block of associated words, determining a plurality of vectors representative of the plurality of words, processing the plurality of vectors using a first neural network to obtain a plurality of recalculated vectors having values based on the plurality of vectors, determining a plurality of association values corresponding to a connections between at least two words of the document, and identifying, using the plurality of recalculated vectors and the plurality of association values, the first block of associated symbol sequences.

Abstract: Systems and methods for optical character recognition using specialized confidence functions. An example method comprises: receiving a grapheme image; computing a feature vector representing the grapheme image in a space of image features; and computing a confidence vector associated with the grapheme image, wherein each element of the confidence vector reflects a distance, in the space of image features, between the feature vector and a center of a class of a set of classes.

Abstract: Systems and methods for neural network training utilizing loss functions reflecting neighbor token dependencies.

Abstract: Systems and methods for performing OCR of an image depicting text symbols and imaging a document having a plurality of planar regions are disclosed. An example method comprises: receiving a first image of a document having a plurality of planar regions and one or more second images of the document; identifying a plurality of coordinate transformations corresponding to each of the planar regions of the first image of the document; identifying, using the plurality of coordinate transformations, a cluster of symbol sequences of the text in the first image and in the one or more second images; and producing a resulting OCR text comprising a median symbol sequence for the cluster of symbol sequences.

Abstract: An example method of detecting repetitive patterns of user interface actions comprises: defining a set of overlapping shingles on a sequence of user interface events; grouping the shingles into a plurality of shingle clusters based on a chosen shingle similarity metric; selecting a shingle cluster having a maximum, among the plurality of shingle clusters, value of a chosen intra-shingle similarity metric; and identifying a repetitive user interface operation represented by the selected shingle cluster.

Abstract: Aspects of the disclosure provide for mechanisms for identification of objects in images using neural networks. A method of the disclosure includes: obtaining an image, representing each element of a plurality of elements of the image via an input vector of a plurality of input vectors, each input vector having one or more parameters pertaining to visual appearance of a respective element of the image, providing the plurality of input vectors to a first subnetwork of a neural network to obtain a plurality of output vectors, wherein each of the plurality of output vectors is associated with an element of the image, identifying, based on the plurality of output vectors, a sub-plurality of elements of the image as belonging to the image of the object, and determining, based on locations of the sub-plurality of elements, a location of an image of an object within the image.

Abstract: Embodiments of the present disclosure describe a system and method for optical character recognition. In one embodiment, a system receives an image depicting text. The system extracts features from the image using a feature extractor. The system applies a first decoder to the features to generate a first intermediary output. The system applies a second decoder to the features to generate a second intermediary output, wherein the feature extractor is common to the first decoder and the second decoder. The system determines a first quality metric value for the first intermediary output and a second quality metric value for the second intermediary output based on a language model. Responsive to determining that the first quality metric value is greater than the second quality metric value, the system selects the first intermediary output to represent the text.

Abstract: Techniques for machine-based identification of objects extracted from text documents in natural language are disclosed. An example method may comprise: identifying matching pairs of one or more information objects corresponding to a real world object, one information object from the document and at least one information object from the document storage for a combination of global identification patterns that exist in the document and in the document storage; ascertaining consistency of the matching pairs and determining which of the one or more information objects in the document are suitable for merging into the document storage; and adding the one or more information objects from the document to the document storage to associate information objects corresponding to the real world object.